Thomas Waldura

At Ekimetrics, I partnered with major clients in retail, automotive, and luxury to design and implement impactful projects. My deliveries were focused on :

In parallel with my final-year at the École Centrale, I joined a research team

from the Hospital University Center. The project I worked on aimed at exploring

the potential improvement of auto-immune illness diagnostic thanks to a deep

learning approach.

The main challenges of this project was :

1. Data Cleaning for Deep Learning use : in order to be able to train deep neural networks I had to standardize a 60 000 images database gathered over 5 years. It was a good exercice to get used to computer vision basic manipulations.

2. Choose the right network : my aim was to create an algorithm that could help doctor to double check their diagnostics. I benchmarked several state-of-the-art algorithms (AlexNet, ResNet, VGG, GANs) and tried different approches to maximize the recall (True Positive Rate).

Within the Research Squad of the R&D department I explored many different ways to perform image restoration (blur removal, noise removal and compression artifacts removal). I studied state-of-the-art models such as Generative Adversarial Models, Transformers and Deep Neural Networks. On the other hand I also implemented and train my own architectures to try to achieve good restoration and image quality assessment performances.